Daily Edition · Page 5

THE DAILY INDEX

Technology, AI, software, security, crypto and business

ATTACKS & VULNERABILITIES

TLDR InfoSec

WHATSAPP SECURITY FLAW EXPOSES 3.5B USERS' DATA FROM ‘BASIC PUBLICLY AVAILABLE INFORMATION'

Austrian researchers discovered that all 3.5B WhatsApp users can be enumerated at a rate of 100M users per hour. The researchers used the WhatsApp Web interface to bulk add users and phone numbers and were able to extract phone numbers for all users, profile photos for 57% of users, and profile text for 27%. Meta implemented rate limiting to address this issue, but also stated that the data exposed was basic publicly available information.

NEXT.JS SECURITY TESTING GUIDE FOR BUG HUNTERS AND PENTESTERS

Deepstrike released a detailed security guide covering common pitfalls in Next.js applications, noting that while the framework includes strong default protections against template injection and XSS, developers can introduce risk through unsafe patterns like dangerouslySetInnerHtml or third-party templating engines. The guide highlights additional attack surfaces, including SSRF via the image optimization pipeline or misconfigured server actions.

SHADOWRAY 2.0: ACTIVE GLOBAL CAMPAIGN HIJACKS RAY AI INFRASTRUCTURE INTO SELF-PROPAGATING BOTNET

ShadowRay 2.0 is a malware campaign targeting Ray, a popular open-source AI framework, to hijack powerful computing clusters and convert them into a global, self-propagating botnet. Attackers exploited legitimate orchestration features within Ray, enabling them to orchestrate compute jobs, steal data, launch DDoS attacks, and autonomously spread across organizations. Attackers quickly weaponized unpatched flaws and misconfigurations in Ray clusters that were mistakenly exposed to the internet.

SUPAPWN: HACKING OUR WAY INTO LOVABLE'S OFFICE AND HELPING SECURE SUPABASE

SupaPwn is a multi-stage exploit chain affecting outdated Supabase environments, where weaknesses in privilege controls, host configurations, and cloud credential handling allowed attackers to escalate far beyond expected tenant-level permissions, even reaching infrastructure belonging to other customers. AI tooling accelerated reconnaissance and exploit development, helping researchers validate the chain quickly. Supabase and Lovable patched all affected systems within a day, limiting impact to a small set of outdated deployments.

SERVICENOW AI AGENTS CAN BE TRICKED INTO ACTING AGAINST EACH OTHER VIA SECOND-ORDER PROMPTS

ServiceNow's Now Assist AI agents can be manipulated through second-order prompt injection, where instructions passed between agents trigger unintended actions. Attackers can use this to escalate privileges, steal data, or redirect workflows. ServiceNow claims the behavior matches expected agent chaining, but organizations are urged to monitor agent interactions and harden guardrails.

TLDR IndexPage 5